An assisted assembly method based on augmented reality

Aiming at the problem of object occlusion in the traditional assembly process, this paper proposes an assisted assembly method based on augmented reality. First, the virtual target model and the on-site assembly scene are used to achieve the initial three-dimensional positioning of the occluded object. Then, an area division method based on bounding box mapping is proposed to efficiently partition the two-dimensional target area required for follow-up tracking registration. Then, a keyframe-based LK optical flow tracking registration method is proposed for real-time pose estimation. This method combines feature matching with the tracking method and realizes smooth switching between them through the tracking failure judgment mechanism. In order to improve the accuracy of tracking registration, a forward and backward error screening mechanism based on the tracking of feature points is added in the feature tracking stage. In the feature matching stage, quadtree homogenization feature extraction and PROSAC mismatching elimination are used. In addition, a jitter correction method based on motion consistency is proposed to further ensure the stability of the algorithm. Finally, according to the tracking registration results, the virtual auxiliary information is fused and superimposed with the actual assembly operation scene, which brings the "perspective" vision beyond reality to the construction personnel and assists the construction personnel in completing the assembly task simply and efficiently. Experimental results show that the proposed method has high tracking accuracy and good stability and can reach 28 frames per second, which can meet the requirements of augmented reality-assisted assembly.